A method for partitioning of input vectors for coding is presented. The method comprises obtaining of an input vector. The input vector is segmented, in a non-recursive manner, into an integer number, NSEG, of input vector segments. A representation of a respective relative energy difference between parts of the input vector on each side of each boundary between the input vector segments is determined, in a recursive manner. The input vector segments and the representations of the relative energy differences are provided for individual coding. Partitioning units and computer programs for partitioning of input vectors for coding, as well as positional encoders, are presented.
Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. An audio coding method, the method comprising: obtaining an input vector of coefficients originating from an audio signal; segmenting the input vector into an integer number (N SEG ) of input vector segments according to a ratio between a total bit-budget for quantizing the input vector and a maximum number of bits allowed for quantizing a vector segment, wherein the maximum number of bits is constrained by a vector quantizer; determining a representation of a respective relative energy difference between parts of the input vector on each side of each boundary between the input vector segments, wherein the determining comprises: a) setting the input vector as an upper level input vector; b) splitting the upper level input vector into left and right parts, each part comprising one or more input vector segments, wherein the upper level input vector is split at the segment boundary between the left and the right part into two lower level input vectors; c) calculating a representation of a relative energy difference between the two lower level input vectors according to an energy ratio between the lower level input vectors; and d) repeating the steps b) and c) of splitting and calculating by re-setting the lower level input vectors as a respective upper level input vector, until all boundaries between input vector segments are provided with an associated representation of a relative energy difference; allocating bits for encoding the shape of each of the input vector segment and for encoding of the representations of the relative energy differences between the input vector segments, wherein bits for encoding the input vector segments are distributed between segments according to relative energy differences between parts of the input vector; and providing each the input vector segment, the representations of the relative energy differences, and allocation information to the vector quantizer for individual encoding of the input vector segments.
Audio coding and compression. This invention addresses the problem of efficiently allocating bits for encoding audio signal coefficients, particularly in vector quantization, by intelligently segmenting the input and representing energy differences between segments. The method begins by obtaining an input vector of audio signal coefficients. This input vector is then segmented into a predetermined number of segments. The number of segments is determined by the total bit budget available for quantization and the maximum number of bits allowed for a single segment, which is limited by the capabilities of a vector quantizer. A representation of the relative energy difference between adjacent parts of the input vector, on either side of each segment boundary, is calculated. This is achieved through a recursive process. The input vector is treated as an upper-level vector and split into left and right parts, each containing one or more segments. The relative energy difference between these two parts is computed, often as an energy ratio. This splitting and calculating process is repeated on the resulting lower-level vectors until all boundaries between the original segments have an associated relative energy difference representation. Subsequently, bits are allocated for encoding the shape of each input vector segment and for encoding these calculated relative energy differences. The distribution of bits allocated to the segments themselves is influenced by the relative energy differences between different parts of the input vector. Finally, the segmented input vector, the representations of the relative energy differences, and the bit allocation information are provided to the vector quantizer for individual encoding of each segment.
2. The method of claim 1 , wherein N SEG is the smallest integer number by which each the input vector segment fulfils constraints associated with a quantizer for the encoding.
This invention relates to vector quantization techniques used in data compression, particularly for optimizing the segmentation of input vectors to meet quantizer constraints. The problem addressed is efficiently determining the smallest integer segment size (N_SEG) that ensures each segment of an input vector satisfies the constraints of a quantizer during encoding. The quantizer imposes specific conditions, such as bitrate limits or distortion thresholds, which must be met for accurate and efficient data representation. The method involves analyzing the input vector to identify the minimal segment size that allows each segment to comply with these constraints without unnecessary division, thereby balancing computational efficiency and encoding quality. This approach is particularly useful in applications like speech coding, image compression, or machine learning where precise quantization is critical. The solution dynamically adjusts the segmentation process based on the quantizer's requirements, ensuring optimal performance across varying input data characteristics. By avoiding overly fragmented segments, the method reduces processing overhead while maintaining high-fidelity encoding. The technique is applicable in systems where input vectors are processed in segments, and the quantizer's constraints must be strictly adhered to for effective compression or representation.
3. The method of claim 1 , wherein if the upper level input vector has to be divided into non-equally sized lower level input vectors, selecting the segment boundary as the boundary closest to the center of the upper level input vector giving a larger last lower level input vector than first lower level input vector.
This invention relates to a method for dividing an upper level input vector into non-equally sized lower level input vectors, particularly in systems where such division is necessary for processing or analysis. The problem addressed is ensuring an optimal segmentation strategy when the input vector cannot be split into equal parts, which is common in applications like signal processing, data compression, or machine learning where fixed-size divisions may not be feasible or efficient. The method involves selecting a segment boundary that is closest to the center of the upper level input vector. This approach ensures that the resulting lower level input vectors are not equally sized but are divided in a way that the last lower level input vector is larger than the first. This asymmetric division can be advantageous in scenarios where the latter part of the input vector contains more significant or complex data, requiring a larger segment for accurate processing. The method avoids arbitrary or inefficient splits by prioritizing a center-aligned boundary, which helps maintain balance while accommodating the need for unequal segmentation. This technique is particularly useful in hierarchical processing systems where different levels of granularity are required for subsequent operations.
4. The method of claim 1 , wherein the step of allocating bits is performed in connection to the step of determining.
A system and method for efficient bit allocation in data encoding processes addresses the challenge of optimizing resource utilization in communication systems. The invention focuses on dynamically assigning bits to data elements based on their significance, improving encoding efficiency and reducing transmission overhead. The method involves determining the importance or priority of data elements, such as in video or audio compression, and then allocating bits accordingly. This allocation step is directly linked to the determination process, ensuring that higher-priority elements receive more bits, while lower-priority elements receive fewer, thereby balancing quality and bandwidth usage. The system may include a processor configured to analyze data characteristics, such as frequency, amplitude, or spatial-temporal relationships, to assess significance. The bit allocation is performed in real-time or near-real-time, adapting to changing data conditions. This approach enhances compression ratios and reduces latency in applications like streaming, storage, and real-time communication. The invention is particularly useful in environments where bandwidth is limited or where high-quality data transmission is critical. By dynamically adjusting bit allocation based on data significance, the system achieves superior performance compared to static allocation methods.
5. The method of claim 1 , wherein the step of allocating bits for encoding of each the input vector segments performed in connection to the step d) calculating a representation of a relative energy difference.
This invention relates to efficient bit allocation for encoding vector segments in signal processing, particularly for audio or speech compression. The problem addressed is optimizing bit allocation to improve compression efficiency while maintaining signal quality, especially in scenarios where input vectors have varying energy levels. The method involves segmenting an input signal into multiple vector segments and calculating a representation of the relative energy difference between these segments. Based on this energy difference representation, bits are allocated for encoding each segment. The allocation ensures that segments with higher energy differences receive more bits, improving perceptual quality while minimizing bitrate. The energy difference calculation may involve comparing segment energies or computing a logarithmic ratio to normalize variations. The method also includes transforming the input vector segments into a frequency domain representation, such as using a modified discrete cosine transform (MDCT), to facilitate efficient encoding. The transformed segments are then quantized and encoded using the allocated bits. The bit allocation step is dynamically adjusted based on the energy difference representation to adapt to varying signal characteristics. This approach enhances compression efficiency by focusing bit allocation on perceptually significant segments, reducing redundancy, and improving overall encoding performance. The method is particularly useful in audio codecs where maintaining high-quality reconstruction with low bitrates is critical.
6. The method of claim 5 , wherein the step of allocating bits allocates bits for the lower level input vectors in dependence of a ratio between lengths of the lower level input vectors and a ratio between the energies in the lower level input vectors.
This invention relates to a method for allocating bits in a signal processing system, particularly for encoding or compressing input vectors in a hierarchical or multi-level structure. The method addresses the challenge of efficiently distributing available bits among multiple input vectors of varying lengths and energy levels to optimize compression performance while maintaining signal quality. The method involves allocating bits to lower-level input vectors based on two key ratios: the ratio between the lengths of the lower-level input vectors and the ratio between their energies. By considering both the length and energy of the vectors, the method ensures that longer vectors or those with higher energy receive more bits, improving the overall efficiency of the encoding process. This approach is particularly useful in applications such as audio, image, or video compression, where different segments of the signal may have varying importance or complexity. The method may be part of a larger encoding or compression system that processes input vectors at different hierarchical levels, where higher-level vectors are derived from or represent combinations of lower-level vectors. The bit allocation step ensures that the lower-level vectors are encoded with an appropriate number of bits, which can then be used to reconstruct the higher-level vectors accurately. This hierarchical approach allows for more efficient compression and better preservation of signal details.
8. An audio encoder for positional encoding, the audio encoding comprising: an input unit configured to receive an input vector representing an audio signal; a partitioning unit configured to partition input vectors of coefficients originating from the audio signal for positional encoding of shapes of the input vectors; a vector quantizer configured to vector quantize segments of an input vector individually, wherein a maximum number of bits allowed for quantizing a vector segment is constrained by the vector quantizer; and an output unit for an encoded signal, wherein the partitioning unit is configured to segment the input vector into an integer number (N SEG ) of input vector segments according to a ratio between a total bit-budget for quantizing the input vector and the maximum number of bits allowed, the partitioning unit is configured to determine a representation of a respective relative energy difference between parts of the input vector on each side of each boundary between the input vector segments by performing a process that includes: a) setting the input vector as an upper level input vector; b) splitting the upper level input vector into left and right parts, each part comprising one or more input vector segments, wherein the upper level input vector is split at the segment boundary between the left and the right part into two lower level input vectors; c) calculating a representation of a relative energy difference between the two lower level input vectors according to an energy ratio between the lower level input vectors; and d) repeating the steps b) and c) of splitting and calculating by re-setting the lower level input vectors as a respective upper level input vector, until all boundaries between input vector segments are provided with an associated representation of a relative energy difference, the partitioning unit is configured to allocate bits for encoding the shape of each of the input vector segment and for encoding of the representations of the relative energy differences between the input vector segments, wherein bits for encoding the input vector segments are distributed between segments according to relative energy differences between parts of the input vector, and the partitioning unit is configured to provide each the input vector segment, the representations of the relative energy differences and allocation information to the quantizer for individual encoding of the input vector segments.
This invention relates to audio encoding for positional encoding of audio signals. The problem addressed is efficiently encoding the shape of audio signals while managing bit allocation to preserve perceptual quality. The system receives an input vector representing an audio signal and partitions it into segments for quantization. A key feature is the dynamic segmentation of the input vector based on a ratio between the total bit budget and the maximum allowed bits per segment. The partitioning unit splits the input vector into segments and calculates relative energy differences between adjacent segments by recursively dividing the vector into left and right parts, computing energy ratios, and repeating the process until all segment boundaries are processed. The system then allocates bits for encoding each segment's shape and the relative energy differences, distributing bits according to the energy distribution across segments. The quantizer individually encodes each segment, the energy difference representations, and allocation information. This approach optimizes bit usage by prioritizing higher-energy segments while maintaining perceptual fidelity through energy-aware partitioning. The method ensures efficient encoding of audio shapes with constrained bit budgets.
9. The audio encoder of claim 8 , wherein N SEG is the smallest integer number by which each the input vector segment fulfils constraints associated with a quantizer for the encoding.
This invention relates to audio encoding, specifically optimizing the segmentation of input audio data for efficient quantization. The problem addressed is ensuring that each segment of an audio signal meets the constraints of a quantizer, which is a critical step in lossy audio compression. The quantizer imposes specific requirements on the input data, such as bit depth or dynamic range, and improper segmentation can lead to inefficiencies or degraded audio quality. The invention describes an audio encoder that processes an input audio signal by dividing it into multiple vector segments. Each segment is analyzed to determine the smallest integer number of segments (N_SEG) required to ensure that every segment meets the constraints of the quantizer. This ensures that the quantization process is both efficient and accurate, preserving audio quality while minimizing computational overhead. The encoder dynamically adjusts the segmentation based on the quantizer's requirements, allowing for adaptive processing of different audio signals. The quantizer constraints may include bit allocation, dynamic range, or other encoding parameters that must be satisfied for proper quantization. By ensuring that each segment adheres to these constraints, the encoder avoids artifacts and maintains high fidelity in the compressed audio output. This approach is particularly useful in real-time audio encoding applications where computational efficiency and audio quality are critical.
10. The audio encoder of claim 8 , wherein the partitioning unit is configured to, if the upper level input vector has to be divided into non-equally sized lower level input vectors, select the segment boundary as the boundary closest to the center of the upper level input vector giving a larger last lower level input vector than first lower level input vector.
This invention relates to audio encoding, specifically improving the partitioning of input vectors in a hierarchical encoding system. The problem addressed is the efficient division of an upper-level input vector into non-equally sized lower-level input vectors, particularly when unequal partitioning is necessary. Traditional methods may lead to suboptimal encoding efficiency or increased computational complexity. The encoder includes a partitioning unit that dynamically selects segment boundaries for vector division. When unequal partitioning is required, the system prioritizes boundaries closest to the center of the upper-level input vector, ensuring the last lower-level input vector is larger than the first. This approach minimizes distortion and maintains encoding efficiency by balancing vector sizes while preserving perceptual audio quality. The partitioning logic is integrated into a broader encoding pipeline, which may include other processing stages like transformation, quantization, or entropy coding, though these are not the focus of this invention. The key innovation lies in the adaptive boundary selection mechanism, which optimizes vector division for scenarios where equal partitioning is impractical. This method reduces artifacts and improves compression performance without additional computational overhead. The solution is particularly useful in high-efficiency audio codecs where precise vector handling is critical.
11. The audio encoder of claim 8 , wherein the partitioning unit is configured to perform the allocating of bits in connection to the determining, in a recursive manner, of a representation of a respective relative energy difference.
This invention relates to audio encoding, specifically improving bit allocation in audio compression to enhance efficiency and quality. The problem addressed is the need for more accurate and efficient bit allocation in audio encoding, particularly when dealing with varying energy levels across different frequency bands or time segments. Traditional methods may not optimally distribute bits, leading to either wasted resources or degraded audio quality. The invention describes an audio encoder with a partitioning unit that allocates bits based on a recursive determination of relative energy differences. The partitioning unit evaluates the energy distribution across different segments of the audio signal, such as frequency bands or time frames, and adjusts bit allocation accordingly. The recursive process involves iteratively refining the bit allocation by comparing energy differences between segments, ensuring that bits are distributed proportionally to the energy content. This approach allows for more precise and adaptive bit allocation, improving compression efficiency while maintaining or enhancing audio quality. The encoder may also include a transformation unit that converts the audio signal into a frequency domain representation, such as using a modified discrete cosine transform (MDCT), to facilitate energy analysis. The partitioning unit then processes this transformed signal to determine the optimal bit allocation. The recursive method ensures that the bit allocation adapts dynamically to changes in the audio signal, such as transitions between high-energy and low-energy segments. This invention is particularly useful in applications requiring high-quality audio compression, such as streaming, storage, and communication systems.
12. The audio encoder of claim 8 , wherein the partitioning unit is configured to allocate bits for encoding of each the input vector segments performed concurrently to the d) calculating of a representation of a relative energy difference.
This invention relates to audio encoding, specifically improving efficiency in bit allocation for concurrent processing of input vector segments. The problem addressed is optimizing bit allocation during encoding to reduce computational overhead while maintaining audio quality. The encoder includes a partitioning unit that divides input audio data into segments for parallel processing. While encoding these segments, the partitioning unit also calculates a representation of the relative energy difference between segments. This calculation is performed concurrently with the encoding process to dynamically adjust bit allocation based on energy variations, ensuring efficient use of available bits. The encoder further includes a quantization unit that quantizes the encoded segments and a bitstream formatter that organizes the quantized data into a structured output. The concurrent calculation of energy differences allows the encoder to prioritize segments with higher energy variations, improving perceptual quality without increasing processing time. This approach is particularly useful in real-time audio applications where low latency and high efficiency are critical. The invention enhances existing audio encoding techniques by integrating energy-based bit allocation into the parallel processing pipeline, reducing complexity while maintaining or improving audio fidelity.
13. The audio encoder of claim 12 , wherein the partitioning unit is configured to perform the allocating of bits by allocating bits for the lower level input vectors in dependence of a ratio between lengths of the lower level input vectors and a ratio between the energies in the lower level input vectors.
This invention relates to audio encoding, specifically improving bit allocation for lower-level input vectors in a hierarchical encoding system. The problem addressed is inefficient bit allocation in multi-level audio encoding, where longer or higher-energy input vectors may not receive proportionate bit resources, leading to suboptimal compression and quality. The encoder includes a partitioning unit that allocates bits to lower-level input vectors based on two key ratios: the ratio of their lengths and the ratio of their energies. By considering both factors, the encoder ensures that longer vectors and those with higher energy receive more bits, improving perceptual quality and compression efficiency. The partitioning unit operates within a hierarchical encoding framework, where input audio is decomposed into multiple levels, and bit allocation is dynamically adjusted to optimize representation. The invention enhances prior art by introducing a more adaptive bit allocation strategy that accounts for both temporal (length) and amplitude (energy) characteristics of the input vectors. This approach reduces distortion in critical audio segments while maintaining efficient compression. The system is particularly useful in high-quality audio codecs where perceptual fidelity is prioritized.
15. A computer program product comprising a non-transitory computer readable medium storing a computer program comprising instructions, which when executed by at least one processor, cause the at least one processor to perform the method of claim 1 .
A system and method for optimizing data processing in a distributed computing environment addresses inefficiencies in task allocation and resource utilization. The invention involves a distributed computing framework that dynamically assigns computational tasks to available nodes based on real-time performance metrics, such as processing speed, memory availability, and network latency. The system monitors the status of each node and adjusts task distribution to balance workloads, preventing bottlenecks and underutilization. A scheduling algorithm prioritizes tasks based on urgency, dependency, and resource requirements, ensuring optimal execution order. The system also includes a fault-tolerance mechanism that detects node failures and redistributes tasks to maintain system reliability. Additionally, the invention incorporates a data caching mechanism to reduce redundant computations by storing intermediate results for reuse. The framework is designed to integrate with existing distributed computing platforms, enhancing their efficiency without requiring significant architectural changes. The solution improves overall system throughput, reduces processing time, and minimizes resource waste, making it suitable for large-scale data processing applications.
16. An audio encoder for positional encoding, the audio encoding comprising: a memory; and processing circuitry coupled to the memory, wherein the audio encoder is configured to: segment an input vector representing an audio signal into an integer number (N SEG ) of input vector segments according to a ratio between a total bit-budget for quantizing the input vector and a maximum number of bits allowed for quantizing a vector segment; determine a representation of a respective relative energy difference between parts of the input vector on each side of each boundary between the input vector segments by performing a process that includes: a) setting the input vector as an upper level input vector; b) splitting the upper level input vector into left and right parts, each part comprising one or more input vector segments, wherein the upper level input vector is split at the segment boundary between the left and the right part into two lower level input vectors; c) calculating a representation of a relative energy difference between the two lower level input vectors according to an energy ratio between the lower level input vectors; and d) repeating the steps b) and c) of splitting and calculating by re-setting the lower level input vectors as a respective upper level input vector, until all boundaries between input vector segments are provided with an associated representation of a relative energy difference, allocate bits for encoding the shape of each of the input vector segment and for encoding of the representations of the relative energy differences between the input vector segments, wherein bits for encoding the input vector segments are distributed between segments according to relative energy differences between parts of the input vector, and provide each the input vector segment, the representations of the relative energy differences and allocation information to a quantizer for individual encoding of the input vector segments.
This invention relates to audio encoding for positional encoding, specifically improving the efficiency of quantizing audio signals by adaptively segmenting and encoding input vectors based on energy distribution. The problem addressed is the need to optimize bit allocation in audio encoding to preserve perceptual quality while minimizing bitrate, particularly in scenarios where audio signals have varying energy across different segments. The audio encoder segments an input vector representing an audio signal into a fixed number of segments (N SEG) based on a ratio between the total available bits and the maximum bits allowed per segment. The encoder then determines the relative energy differences between adjacent segments by recursively splitting the input vector into left and right parts, calculating energy ratios at each boundary, and repeating the process until all segment boundaries are analyzed. This hierarchical approach ensures that energy differences are accurately captured at multiple levels of granularity. Bits are allocated for encoding both the shape of each segment and the relative energy differences between segments, with bit distribution adjusted according to the energy differences. Higher-energy segments receive more bits, while lower-energy segments receive fewer, optimizing perceptual quality. The segments, energy difference representations, and bit allocation information are then passed to a quantizer for individual encoding. This method improves encoding efficiency by dynamically adapting to the signal's energy distribution.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
June 5, 2020
February 1, 2022
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.